Real-time query system for real estate registration information based on graphic data

Abstract

The invention discloses a real estate registration information real-time query system based on graphic data, and relates to the technical field of information real-time query, and the system comprises a knowledge graph generation unit which is used for processing an established multi-modal data set, generating a joint vector, coding the joint vector, and storing the coded joint vector in a preset graph database, synchronously constructing a dual dynamic index mechanism to generate a real estate knowledge graph; the knowledge graph updating unit is used for capturing change events in the multi-modal data set in real time, identifying graph structure differences of a change area and generating a topology correction strategy by positioning the change area and a neighborhood range, and updating a real estate topology knowledge graph in combination with transaction characteristics of a graph database; a query result display unit; and a data synchronization unit. According to the method, the real-time data channel is constructed, capturing and incremental processing of the real estate business change event are achieved, and it is ensured that the real estate knowledge graph and graphic data are dynamically consistent in topological structure and attribute information.

Description

Technical Field

[0001] This invention relates to the field of real-time information query technology, and more specifically, to a real-time query system for real estate registration information based on graphic data. Background Technology

[0002] With the rapid development of information technology, especially the continuous advancement of geographic information systems and big data technology, the methods for querying real estate registration information have also been significantly improved. Traditional query systems mostly rely on static databases, obtaining information through manual input of conditions and browsing records one by one, resulting in problems such as low query efficiency, susceptibility to errors, and poor interactivity. As real estate information continues to increase, the limitations of traditional query systems become increasingly apparent, especially when dealing with complex spatial data, rapid response, and large-scale queries, where traditional systems prove inadequate. Furthermore, most existing real estate management systems use text-based or tabular formats to display real estate information. While these can meet basic query needs, they fail to fully utilize the advantages of spatial data during user queries, leading to a non-intuitive query process and difficulty in displaying spatial distribution or relationships. For example, in applications such as urban planning and real estate market analysis, users need to be able to quickly query detailed information about specific plots and buildings through visual maps and perform dynamic interactive operations on related geographical locations.

[0003] In recent years, with the widespread application of graphic data (such as vector data and raster data) and spatial data analysis technologies, real estate registration information query systems based on graphic data have gradually become a new research direction. The introduction of graphic data allows for a more accurate description of the spatial location, morphological characteristics, and surrounding environment of real estate, providing users with a more intuitive and interactive query experience. However, existing query systems generally adopt a separate architecture of "attribute library + graphic library." In this architecture, graphic data is usually stored in the form of static images or CAD files, lacking semantic association and synchronous update mechanisms with attribute information. This leads to the need for cross-system calls during spatial queries, increasing system coupling and easily causing data inconsistency issues. Furthermore, real estate registration involves frequent operations such as ownership changes, division and merger, and mortgage seizures. However, most existing systems use batch synchronization or scheduled tasks to update graphic data, failing to detect change events in real time and perform incremental processing. This results in a lag in updating the graphic topology, easily creating data gaps, thus affecting the continuity of registration operations and the effectiveness of supervision.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In response to the problems in related technologies, this invention proposes a real-time query system for real estate registration information based on graphic data, in order to overcome the aforementioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: A real-time query system for real estate registration information based on graphic data, the system comprising: The knowledge graph generation unit is used to process the established multimodal dataset, generate joint vectors, encode and store the joint vectors in a preset graph database, and simultaneously build a dual dynamic indexing mechanism to generate a real estate knowledge graph. The knowledge graph update unit is used to capture change events in the multimodal dataset in real time. By locating the changed area and its neighborhood, it identifies the graph structure differences in the changed area and generates a topology correction strategy. It then updates the real estate topology knowledge graph by combining the transaction characteristics of the graph database. The query results display unit is used to build a semantic parsing mechanism to convert users' natural language queries into corresponding graph query languages ​​and display the query results using visualization tools. The data synchronization unit is used to acquire real estate business operations and convert them into events in a predefined format. Event transmission is performed through a distributed message queue, and real-time processing is combined with streaming topology technology to ensure the consistency between the real estate knowledge graph and the graphical data.

[0007] Furthermore, the knowledge graph generation unit includes: The dataset creation module is used to collect graphic data and related attribute information of real estate from multi-source heterogeneous platforms, and combine spatiotemporal benchmark dynamic alignment algorithm and blockchain-based spatiotemporal anchoring mechanism to create a multimodal dataset. The joint vector generation module is used to construct a dual-channel adversarial generative network that includes geometric and semantic channels, and to perform cross-modal adversarial training on the established multimodal dataset to generate joint vectors. The storage and construction module is used to encode joint vectors into composite semantic codes using a spatiotemporal-aware encoder, and write the semantic codes into a preset graph database, while simultaneously constructing a dual dynamic indexing mechanism of R* tree spatial index and semantic inverted index. The knowledge graph building module is used to analyze the semantic similarity and topological relationships between entity nodes in real time based on the completed graph database and the constructed dual dynamic indexing mechanism, and to build a real estate knowledge graph.

[0008] Furthermore, the storage and building modules include: The composite semantic encoding submodule is used to reconstruct the graph attention mechanism using a pre-trained spatiotemporal awareness encoder, design a multi-layer attention network, generate a fused feature vector, and map the fused feature vector to a composite semantic code through a multi-layer perceptron. The storage submodule is used to select the native stream processing pipeline of the graph database and write the generated composite semantic code and the corresponding multimodal dataset as batch transactions into the preset graph database for storage. The index construction submodule is used to construct an R* tree spatial index based on the spatial grid features in the composite semantic encoding, and to parse the entity type code and relation topology code of the composite semantic encoding to construct a semantic inverted index. The indexing mechanism submodule is used to build a dual dynamic indexing mechanism by sharing storage pointers of composite semantic encoding, so as to perform real-time association and collaborative access between the R* tree space index and the semantic inverted index.

[0009] Furthermore, the composite semantic coding submodule includes: By combining a pre-trained spatiotemporal-aware encoder with a spatial relation distillation mechanism, spatial topological prior vectors are generated, and the graph attention mechanism is reconstructed by combining preset rule constraints. Based on the reconstructed graph attention mechanism, a multi-layer attention network is designed, which includes a spatial grid layer, a temporal slice layer, an entity type layer, and a relation topology layer. The joint vector is input into a multi-layer attention network. By calculating the attention weight of each layer, the corresponding feature vector is output. The feature vectors are then weighted and fused together with the weight matrix of each layer to generate a fused feature vector. By using a multilayer perceptron to perform nonlinear mapping on the fused feature vectors, a composite semantic code expressing spatial location, temporal context, entity category, and topological semantics is generated.

[0010] Furthermore, by combining a pre-trained spatiotemporal-aware encoder with a spatial relation distillation mechanism, spatial topological prior vectors are generated. Then, by incorporating pre-defined rule constraints, the graph attention mechanism is reconstructed, including: Based on the pre-trained spatiotemporal-aware encoder, a hierarchical graph convolutional network is used to extract topological features at different spatial scales, and the topological features are distilled and abstracted to generate spatial topological prior vectors. The preset rule constraints are transformed into rule damage functions, and combined with the generated spatial topology prior vectors, a rule constraint engine containing spatial topology information and rule semantic relationships is constructed. Based on the graph attention mechanism in the spatiotemporal awareness encoder, the graph attention mechanism is reconstructed using the topological priors and rule biases derived from the rule constraint engine.

[0011] Furthermore, the knowledge graph update unit includes: The change event capture module is used to build an event listening pipeline using message flow technology to capture change events of graphical data in multimodal datasets in real time. The change area positioning module is used to locate the change area and its surrounding range using multi-level spatial indexing technology based on the coordinate information of the change event. The graph structure optimization module is used to analyze and identify graph structure differences in changed areas, formulate topology correction strategies using topology correction technology, and optimize the graph structure of changed areas by combining preset business weights and priorities. The knowledge graph update module is used to update the real estate topology knowledge graph based on the optimized graph structure and the transaction characteristics of the graph database.

[0012] Furthermore, the graph structure optimization module includes: The target node pair filtering submodule is used to construct a local subgraph based on the located change area and neighborhood range, and to filter out target node pairs associated with the change area through spatial indexing technology. The graph structure difference identification submodule is used to perform topological relationship calculations on the selected target node pairs using incremental graph algorithms to identify graph structure differences in changed areas. The topology correction policy submodule is used to construct a topology correction generative adversarial network, which combines a reinforcement learning framework with stress field analysis techniques to generate dynamic reward signals and generate topology correction policies based on the dynamic reward signals. The graph structure update submodule is used to verify the topology correction strategy using incremental graph search technology, and combined with preset business weights and priorities, apply the verified topology correction strategy to the changed area to optimize the graph structure of the changed area.

[0013] Furthermore, the topology correction strategy submodule includes: Based on the identified graph structure differences, a topology-modified generative adversarial network containing a generator and a discriminator is constructed. By combining a reinforcement learning framework, the pre-defined business rules are mapped to the physical characteristics of fluids, and dynamic reward signals are generated through pressure field analysis technology to guide the generator to make topology corrections. The discriminator is used to evaluate the rationality of the topology correction results generated by the generator, and the evaluation results are fed back to the generator to optimize the topology correction parameters in order to generate a topology correction strategy.

[0014] Furthermore, by combining a reinforcement learning framework, pre-defined business rules are mapped to the physical characteristics of fluids. Dynamic reward signals are generated through pressure field analysis techniques to guide the generator in topology corrections, including: A reinforcement learning framework is embedded in a topology-modified generative adversarial network, and pre-defined business rules are mapped to the physical characteristics of fluids to construct a fluid dynamics model. A pressure gradient field is generated using a fluid dynamics model, abrupt change points are identified using wavelet transform, and a side length threshold algorithm based on a triangular mesh is used to obtain pressure abrupt change regions, extract spatial heterogeneity features, and generate a dynamic reward signal. The generated dynamic reward signal is used as feedback input in the reinforcement learning framework to guide the generator to make topology corrections during adversarial training.

[0015] Furthermore, a pressure gradient field is generated using a fluid dynamics model, abrupt change points are identified using wavelet transform, and a side-length threshold algorithm based on a triangular mesh is used to obtain pressure abrupt change regions. Spatial heterogeneity features are extracted, and a dynamic reward signal is generated, including: The pressure field distribution is obtained by numerically simulating the target space using a fluid dynamics model, and the spatial gradient of the pressure field is calculated by the finite difference algorithm to generate the pressure gradient field. Wavelet transform was used to perform multi-scale analysis on the generated pressure gradient field, identify abrupt change points, and establish a cluster of abrupt change points. Construct a triangular network based on the established mutation point group, and draw a curve between the cumulative number of sides and the length of the triangle side. Calculate the slope of the initial line segment connecting the first and last points in the curve. The curves sorted by triangle side lengths are segmented using a differential algorithm. The slope of the line segment connecting the first and last points of each curve segment is compared with the slope of the initial line segment. A threshold is set based on the comparison results to filter out regions of sudden pressure changes. The pressure gradient field and pressure change region are analyzed to establish the physical relationship between fluid pressure and spatial topological changes, and spatial non-uniformity analysis is performed to extract spatial heterogeneity characteristics. The extracted spatial heterogeneity features are transformed into reward values ​​through nonlinear exponential mapping, and combined with random exploration incentive terms to generate dynamic reward signals.

[0016] The beneficial effects of this invention are as follows: 1. This invention introduces distributed message queues and streaming topology technology to build a real-time data channel, enabling the capture and incremental processing of real estate business change events. This avoids the data lag problem caused by traditional batch synchronization, ensures the dynamic consistency of real estate knowledge graphs and graphical data in terms of topological structure and attribute information, and improves the system response speed and the accuracy and timeliness of data synchronization.

[0017] 2. This invention achieves the storage and accurate querying of massive real estate data through joint vector coding and dual dynamic indexing mechanism, improves the processing performance of graph database, ensures rapid response of data retrieval and update, and enhances the stability and scalability of the system in large-scale data environment.

[0018] 3. This invention enables users to query real estate information in natural language through semantic parsing mechanisms and visualization tools, reducing the operational threshold and improving the interactive experience; at the same time, the generated topology correction strategy ensures that the graph structure continues to conform to the actual business situation, enhancing the system's adaptability and intelligent decision-making capabilities in dynamic environments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a real-time query system for real estate registration information based on graphic data according to an embodiment of the present invention;

[0021] Figure 2 This is a data fusion and functional design path diagram of a real-time query system for real estate registration information based on graphic data according to an embodiment of the present invention; Figure 3 This is an architecture diagram of a real-time query system for real estate registration information based on graphical data, according to an embodiment of the present invention.

[0022] In the picture: 1. Knowledge graph generation unit; 2. Knowledge graph update unit; 3. Query result display unit; 4. Data synchronization unit. Detailed Implementation

[0023] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0024] According to an embodiment of the present invention, a real-time query system for real estate registration information based on graphic data is provided.

[0025] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a real-time query system for real estate registration information based on graphic data includes: The knowledge graph generation unit 1 is used to process the established multimodal dataset, generate joint vectors, encode and store the joint vectors in a preset graph database, and simultaneously build a dual dynamic indexing mechanism to generate a real estate knowledge graph.

[0026] In this optional embodiment, the knowledge graph generation unit 1 includes: The dataset creation module is used to collect graphic data and associated attribute information of real estate from multi-source heterogeneous platforms, and to create a multimodal dataset by combining a spatiotemporal benchmark dynamic alignment algorithm and a blockchain-based spatiotemporal anchoring mechanism.

[0027] It should be further explained that the process of collecting graphic data and associated attribute information of real estate from multi-source heterogeneous platforms, and combining it with a spatiotemporal benchmark dynamic alignment algorithm and a blockchain-based spatiotemporal anchoring mechanism to establish a multimodal dataset specifically includes: First, real estate graphic data (such as land boundaries, building outlines, 3D models, etc.) and their related attribute information (such as property rights, use, area, geographic coordinates, etc.) are collected from various heterogeneous data sources (such as GIS platforms, CAD files, real estate databases, etc.) via interfaces or APIs. Next, a spatiotemporal benchmark dynamic alignment algorithm is applied to adjust the alignment of data from different sources based on time and space factors, thereby ensuring data consistency and accuracy. At this point, all data undergoes timestamp processing to ensure that data from different sources are synchronized at the same point in time. Simultaneously, a blockchain-based spatiotemporal anchoring mechanism is used to leverage the blockchain's... The decentralized nature and immutability of blockchain provide reliable verification and protection for the spatiotemporal positioning of data. By storing spatiotemporally anchored information through blockchain, it ensures higher security and transparency of real estate data during updates, queries, and sharing. Finally, it forms a multimodal dataset, which includes not only graphical data but also attribute data, and all data can be accurately mapped to specific spatiotemporal points, providing unified and accurate resources to support further analysis and real-time queries. At the same time, it reduces the data coupling between systems and improves the timeliness and credibility of data, which is of great significance for real estate registration, changes, queries, and supervision.

[0028] In a specific embodiment, the dataset creation module analyzes real estate registration information, exports spatial and non-spatial data that can be publicly displayed offline, and uses a non-linear confidentiality processing algorithm to declassify spatial data such as cadastral areas, cadastral sub-areas, parcels, and natural buildings in the real estate cadastral data, and desensitizes the real estate registration data to strictly protect the privacy information of the rights holders, filter out sensitive information related to the rights holders, and ensure that personal information is not disclosed.

[0029] like Figure 2As shown, by integrating spatial data such as cadastral districts, cadastral sub-districts, parcels, and natural buildings from the organized real estate registration database, a natural building model is constructed. Through the building list interface, the registration information associated with real estate units can be retrieved in real time. The basic information of any parcel, natural building list information, natural attributes of the house, and restricted rights can be viewed online, realizing the "house search by map" function. At the same time, parcel maps and house sub-unit maps can be viewed and downloaded, which innovates the query method, broadens the query channels, and saves query costs.

[0030] The joint vector generation module is used to construct a dual-channel adversarial generative network that includes geometric and semantic channels, and to perform cross-modal adversarial training on the established multimodal dataset to generate joint vectors.

[0031] It should be further explained that in the dual-channel adversarial generative network, the geometric channel is responsible for processing the graphic data of real estate (such as land boundaries, building shapes, etc.), extracting the spatial features of the graphic data through a convolutional neural network; the semantic channel focuses on processing the attribute information of real estate (such as ownership, use, area, etc.), extracting the deep semantic features of the attribute data through a fully connected network or transformer model. Furthermore, during the training process of the adversarial generative network, the generator generates graphic data and attribute data in the geometric and semantic channels respectively, while the discriminator evaluates whether the generated data is consistent with the real data and optimizes the generator through feedback. To ensure the uniformity and relevance of cross-modal data, a cross-modal adversarial training strategy is adopted to ensure that geometric data and semantic data can effectively complement each other during the generation process, thereby generating a joint vector that simultaneously contains graphic and attribute information, providing unified data support for subsequent queries, analysis, and decision-making.

[0032] The storage and construction module is used to encode joint vectors into composite semantic codes using a spatiotemporal-aware encoder, and write the semantic codes into a preset graph database, while simultaneously constructing a dual dynamic indexing mechanism of R* tree spatial index and semantic inverted index.

[0033] In this optional embodiment, the storage and build module includes: The composite semantic encoding submodule is used to reconstruct the graph attention mechanism using a pre-trained spatiotemporal awareness encoder, design a multi-layer attention network, generate a fused feature vector, and map the fused feature vector into a composite semantic code through a multi-layer perceptron.

[0034] In this optional embodiment, the composite semantic coding submodule includes: By combining a pre-trained spatiotemporal-aware encoder with a spatial relation distillation mechanism, spatial topological prior vectors are generated, and the graph attention mechanism is reconstructed by combining preset rule constraints.

[0035] In this optional embodiment, a spatial topological prior vector is generated by combining a pre-trained spatiotemporal awareness encoder with a spatial relation distillation mechanism, and the graph attention mechanism is reconstructed by combining preset rule constraints, including: Based on the pre-trained spatiotemporal-aware encoder, a hierarchical graph convolutional network is used to extract topological features at different spatial scales, and the topological features are distilled and abstracted to generate spatial topological prior vectors. The preset rule constraints are transformed into rule damage functions, and combined with the generated spatial topology prior vectors, a rule constraint engine containing spatial topology information and rule semantic relationships is constructed. Based on the graph attention mechanism in the spatiotemporal awareness encoder, the graph attention mechanism is reconstructed using the topological priors and rule biases derived from the rule constraint engine.

[0036] Based on the reconstructed graph attention mechanism, a multi-layer attention network is designed, which includes a spatial grid layer, a temporal slice layer, an entity type layer, and a relation topology layer.

[0037] The joint vector is input into a multi-layer attention network. By calculating the attention weight of each layer, the corresponding feature vector is output. The feature vectors are then weighted and fused together with the weight matrix of each layer to generate a fused feature vector.

[0038] By using a multilayer perceptron to perform nonlinear mapping on the fused feature vectors, a composite semantic code expressing spatial location, temporal context, entity category, and topological semantics is generated.

[0039] The storage submodule is used to select the native stream processing pipeline of the graph database and write the generated composite semantic code and the corresponding multimodal dataset as batch transactions into the preset graph database for storage.

[0040] The index construction submodule is used to construct an R* tree spatial index based on the spatial grid features in the composite semantic encoding, and to parse the entity type code and relation topology code of the composite semantic encoding to construct a semantic inverted index.

[0041] It should be further explained that, based on the spatial grid features in the composite semantic encoding, an R* tree spatial index is constructed, and the entity type code and relation topology code of the composite semantic encoding are parsed to construct a semantic inverted index, including: First, based on the spatial grid features in the composite semantic encoding, an R* tree spatial index is constructed. The R tree is an adaptive multidimensional data structure suitable for storing and retrieving spatial data. By organizing spatial objects (such as plots, buildings, etc.) into a tree structure according to their geometry and spatial location, it can effectively support spatial operations such as region queries and neighborhood queries. When constructing the R tree, each spatial object is mapped to a rectangular bounding box, and indexing is performed by recursively allocating spatial regions to improve query efficiency and reduce computational complexity.

[0042] Next, by parsing the entity type code and relation topology code in the composite semantic code, the meaning of the composite semantic code is converted into actual spatial and semantic information; the entity type code represents different objects in space (such as residential buildings, commercial buildings, etc.), while the relation topology code describes the relationships between these objects (such as proximity relationships, connection relationships, etc.).

[0043] Finally, a semantic inverted index is constructed, which maps each entity type and its corresponding relational topology to the index of a document or dataset, thereby enabling efficient retrieval of content related to specific entity types and spatial relationships. The semantic inverted index supports fast querying and retrieval by recording the occurrence position of each entity.

[0044] The indexing mechanism submodule is used to build a dual dynamic indexing mechanism by sharing storage pointers of composite semantic encoding, so as to perform real-time association and collaborative access between the R* tree space index and the semantic inverted index.

[0045] The knowledge graph building module is used to analyze the semantic similarity and topological relationships between entity nodes in real time based on the completed graph database and the constructed dual dynamic indexing mechanism, and to build a real estate knowledge graph.

[0046] It should be further explained that, based on the completed graph database and the constructed dual dynamic indexing mechanism, the semantic similarity and topological relationships between entity nodes are analyzed in real time to establish a real estate knowledge graph, specifically including: Firstly, by utilizing the completed graph database and the constructed dual dynamic indexing mechanism, spatial and semantic information related to real estate can be efficiently stored and accessed. The graph database provides a storage structure of nodes and edges, where nodes represent real estate entities (such as land parcels, buildings, areas, etc.) and edges represent the relationships between these entities (such as adjacency, inclusion, connection, etc.). Through the dual dynamic indexing mechanism, spatial data (such as location, boundaries, etc.) and semantic data (such as use, ownership, etc.) can be correlated and accessed in real time during queries, ensuring the accuracy and efficiency of queries.

[0047] Next, by analyzing the semantic similarity and topological relationships between entity nodes in real time, the nodes in the graph database can be analyzed to identify entity nodes of the same or similar types, and the relationships between them can be determined based on spatial location and semantic information. Semantic similarity analysis is usually based on semantic embedding technology, which judges the similarity between nodes by comparing the attribute information of nodes (such as building type, area, years of use, etc.). Topological relationship analysis is based on R* tree spatial index, which analyzes the spatial adjacency and connectivity between nodes to capture the spatial structural relationships between different entities.

[0048] Finally, based on these analysis results, a real estate knowledge graph is established. This graph will include real estate entities (nodes), semantic relationships between entities (such as use, ownership, etc.), spatial topological relationships (such as adjacency, distance, etc.), and entity attribute information, which can help users make more accurate queries and decisions.

[0049] It should be noted that although the R* tree spatial index is used to capture the spatial structural relationships between different entities, users do not care about all the details, especially the specific form of the graphic data, during the rendering process. In some embodiments, the rendering module can distinguish the specific form of the graphic data based on the attribute information selected by the user. In this paragraph, the attribute information may also include: the floor height of the community, the shortest distance to key areas (e.g., schools, hospitals, etc.), the unit price per square meter, etc.; more specifically, the specific value of the selected attribute information can be corresponding to the depth of a specified color: for example, the larger the specific value of the selected attribute information, the darker the specified color; otherwise, the lighter the specified color.

[0050] It's important to note that when graphics data uses an R-tree spatial index as its underlying storage structure, directly using user-selected attribute information as the driving condition for rendering often fails to meet the demands for high real-time performance and high responsiveness. This is because the core advantage of the traditional R-tree index structure lies in spatial range queries and the hierarchical organization of geometric objects. Its design goal is primarily focused on the storage and retrieval efficiency of spatial entities, rather than optimization for the high-frequency refresh scenarios of interactive rendering. During multiple interactive operations, such as frequent dragging, zooming, or switching of attribute information, the node retrieval and update operations of the R-tree need to be constantly split, merged, or relocated. This leads to significant performance overhead and response latency, making it difficult to meet the millisecond-level feedback requirements of real-time rendering.

[0051] Furthermore, if the system redraws all graphical data for every attribute switch or interactive operation, it not only requires retracing the R-tree nodes and generating the complete pixel matrix, but also reassigning and redrawing every element of the graphic. This full redraw significantly increases computational resource consumption and causes noticeable lag and stuttering during user interface operations, severely impacting system smoothness and user experience. In large-scale data scenarios, such as visualization of tens of thousands of community units or complex geographical areas, the cost of full redrawing is particularly high, making it almost impossible to promote in practical applications.

[0052] Therefore, in some embodiments, this application does not employ the traditional "clear canvas - redraw" approach, but instead proposes an overlay rendering mechanism. This mechanism incrementally updates the rendering layer only for the user-selected attribute, based on the existing graphical results. This effectively avoids redundant calculations of unchanged parts, achieving a rendering logic of "static part reuse, dynamic part overlay," thereby significantly reducing the performance overhead caused by repeated drawing. In some embodiments, the query result display unit further includes a rendering module, which can render the graphical data of the selected attribute information, specifically: Obtain the base coordinates of the current rendering interface, and extract the bounding box of the rectangle under the current rendering interface based on the R* tree space index.

[0053] It is understandable that the rectangular bounding box contains the associated information of each node.

[0054] In response to the selected attribute information, and based on the bounding box of the current rendering interface, update the rendering mapping information of each node in the bounding box.

[0055] It is understandable that the larger the specific value of the node's attribute information, the larger the rendering mapping information of the updated node; otherwise, the smaller the rendering mapping information.

[0056] In order to balance rendering real-time performance and smooth scrolling, after each response to the selected attribute information, it is necessary to generate the graphics data matrix A of the current rendering interface.

[0057] It is understandable that the graphics data matrix A is a pixel matrix, which refers to the rendering interface before the attribute information is selected. Each element a_ij represents the pixel value of the pixel in the i-th row and j-th column of the rendering interface. This pixel value can be a vector, for example: a_ij=[l_ij, c_ij, w_ij], where l_ij, c_ij, and w_ij represent the brightness, hue, and saturation of the pixel, respectively.

[0058] In response to user movement of the rendering interface, the system uses a graphics data matrix A as the underlying logic for switching rendering, overlaying the rendering mapping information of nodes onto the graphics data matrix. When the user performs operations such as sliding, dragging, or zooming on the rendering interface, the system needs to ensure that the screen refreshes in real time and maintains a smooth interactive experience. Therefore, in some embodiments, the graphics data matrix A can be used as the underlying logic for switching rendering. Specifically, when the user triggers screen movement, the system first obtains the new viewport range and compares it with the previous rendering range to determine which areas have moved out of the viewport and which areas are newly entered. Subsequently, the system does not re-render the entire screen, but instead quickly retrieves the bounding box nodes associated with the new viewport area by calling the R* tree index structure, and processes only the graphics data corresponding to these nodes. For parts still within the viewport range that have not changed, the system directly reuses the pixel information stored in matrix A as a cache base, thus avoiding large-scale redrawing. For newly entered areas of the viewport, the system quickly generates the corresponding rendering layer based on the attribute information carried by the nodes and combined with pre-defined color mapping rules, and overlays and updates it on matrix A. In this way, matrix A acts as a sliding cache during interface transitions: it retains the rendering results of unchanged areas while supporting incremental updates to new areas. Finally, the incrementally rendered matrix A is submitted to the display module, synchronized with the user's sliding actions, ensuring smooth interface refresh without delays or blank displays during movement. Compared to traditional full redrawing, this process significantly reduces system computational overhead and effectively improves responsiveness and smoothness in user interaction, thus meeting high real-time requirements while maintaining rendering accuracy.

[0059] Through this improvement, the system can maintain the advantages of spatial index structure in storage and retrieval, while introducing caching and overlay mechanisms during the rendering process. This balances the rigor of data management with the high responsiveness of interface rendering, providing effective support for the real-time visualization of large-scale graphic data.

[0060] However, the dynamic overlay update method also has certain drawbacks, mainly that the interface is not smooth enough during the switching process, especially when switching between selected attribute information on the same interface, the transition is prone to be unnatural.

[0061] In some embodiments, updating the rendering mapping information of each node specifically includes: the rendering mapping information is based on the HSV model; and a hue reference matrix is ​​pre-generated.

[0062] The hue reference matrix [xij] is an n*n matrix. Each attribute, including the first and second attributes mentioned below, corresponds to a specific color h1, h2, ..., hn, where n is the number of optional attribute information. Therefore, xij refers to the preset saturation difference when switching from color hi to color hj to reduce visual abruptness.

[0063] Understandably, when colors change, hue actually changes, and color purity corresponds to saturation.

[0064] Obtain the first attribute information and its corresponding saturation vector from the current rendering interface. In response to selecting the second attribute information, find the corresponding saturation difference in the hue reference matrix and add it to the saturation vector to obtain the target vector. That is, each element of the target vector is the corresponding element in the saturation vector plus the saturation difference. Substitute the minimum distinguishing value, the target vector, and the specific values ​​of the second attribute information for each node in the bounding box into the dynamic programming algorithm to obtain the saturation vector corresponding to the second attribute information.

[0065] Understandably, in rendering mapping information, the brightness value generally changes accordingly with the saturation, and the hue is determined by the selected attribute information. Therefore, as long as the selected attribute information, such as the saturation vector corresponding to the second attribute information, is determined, the rendering mapping information is also determined. The minimum distinguishing value refers to the minimum effective interval value that saturation needs to have under the same hue condition to ensure that the human eye can reliably distinguish adjacent color levels.

[0066] Understandably, in this dynamic programming algorithm, its time complexity directly depends on the number of calculations, i.e., the quotient of the range of saturation values ​​and the minimum discriminant value. Furthermore, its optimization objective is min{SSD(s,S)}, where SSD is the sum of squared differences function, and s and S represent the target vector and the saturation vector corresponding to the second attribute information, respectively. In addition, the constraint is that the saturation vector corresponding to the second attribute information should be linearly related to the specific values ​​of the second attribute information of each node in the bounding box.

[0067] Knowledge graph update unit 2 is used to capture change events in the multimodal dataset in real time. By locating the changed area and its neighborhood, it identifies the graph structure differences in the changed area and generates a topology correction strategy. It then updates the real estate topology knowledge graph by combining the transaction characteristics of the graph database.

[0068] In this optional embodiment, the knowledge graph update unit 2 includes: The change event capture module is used to build an event listening pipeline using message flow technology to capture change events of graphical data in multimodal datasets in real time.

[0069] It should be further explained that the use of message flow technology to build an event listening pipeline to capture change events of graphical data in multimodal datasets in real time specifically includes: By leveraging message streaming technologies (such as Kafka and RabbitMQ) to build an event listening pipeline, we can capture any events related to graphic data by monitoring sources of change in the graphic data (such as change logs in the graph database and update notifications from the GIS system). Whenever the data changes, the message streaming system packages the relevant event data into a message and pushes it into the pipeline. The message contains information such as the change type (e.g., addition, modification, deletion), the change content (e.g., updated coordinates, attributes), and the timestamp of the event. These messages are received and processed in real time by consumers in the pipeline (such as processing engines and data synchronization systems), thereby enabling timely perception and response to changes in graphic data.

[0070] The change area positioning module is used to locate the change area and its surrounding range using multi-level spatial indexing technology based on the coordinate information of the change event.

[0071] It should be further explained that, based on the coordinate information of the change event, the multi-level spatial indexing technology is used to locate the changed area and its surrounding range, specifically including: First, multi-dimensional spatial indexing techniques such as R* trees are used to manage spatial data hierarchically. R* trees divide spatial data into multiple small regions and improve spatial query efficiency through recursive partitioning. When a change event is received, the system quickly searches through the R* tree based on the coordinates of the change to determine the specific region where the changed data is located. Then, through multi-level spatial indexing techniques, combined with indexes of different spatial levels (such as using coarse-grained global region indexes and fine-grained local region indexes), the precise location of the changed region can be accurately located. At the same time, based on the coordinates of the changed region, the system further queries and determines the neighborhood range, that is, the related spatial entities within a certain distance around the changed region, thereby improving the efficiency of change event processing and ensuring the rapid location of the changed region and its impact range in large-scale spatial data.

[0072] The graph structure optimization module is used to analyze and identify graph structure differences in changed areas, formulate topology correction strategies using topology correction technology, and optimize the graph structure of changed areas by combining preset business weights and priorities.

[0073] In this optional embodiment, the graph structure optimization module includes: The target node pair filtering submodule is used to construct a local subgraph based on the located change area and its neighborhood range, and to filter out target node pairs associated with the change area using spatial indexing technology.

[0074] It should be further explained that, based on the location change area and its neighborhood, a local subgraph is constructed, and spatial indexing technology is used to filter out target node pairs associated with the change area, specifically including: Based on the spatial relationships of the changed area, all spatial entities (such as plots of land, buildings, roads, etc.) within the area are extracted, and a local subgraph is constructed using these entities as nodes. The local subgraph contains all target nodes that are directly or indirectly connected to the changed area, including neighboring nodes of the changed area and related nodes that may be affected by the change, such as adjacent plots of land and connected buildings. Target node pairs associated with the changed area are filtered out using spatial indexing technology, that is, those node pairs with close spatial or semantic relationships. The filtering process uses distance calculation or neighborhood lookup in the spatial index to ensure that only node pairs affected by the change event are included, thereby avoiding interference from irrelevant nodes. This not only ensures the construction of the local subgraph and the accurate filtering of target node pairs, but also significantly improves query and computation efficiency.

[0075] The graph structure difference identification submodule is used to perform topological relationship calculations on the selected target node pairs using incremental graph algorithms to identify graph structure differences in changed areas.

[0076] The topology correction policy submodule is used to construct a topology correction generative adversarial network, which combines a reinforcement learning framework with stress field analysis techniques to generate dynamic reward signals and generate topology correction policies based on the dynamic reward signals.

[0077] In this optional embodiment, the topology correction strategy submodule includes: Based on the identified differences in graph structure, a topology-modified generative adversarial network containing a generator and a discriminator is constructed.

[0078] By combining a reinforcement learning framework, pre-defined business rules are mapped to the physical characteristics of fluids, and dynamic reward signals are generated through pressure field analysis technology to guide the generator to perform topology correction.

[0079] In this optional embodiment, by combining a reinforcement learning framework, preset business rules are mapped to the physical characteristics of a fluid, and dynamic reward signals are generated through pressure field analysis technology to guide the generator in topology correction, including: A reinforcement learning framework is embedded in a topology-modified generative adversarial network, and pre-defined business rules are mapped to the physical characteristics of fluids to construct a fluid dynamics model.

[0080] It should be further explained that embedding a reinforcement learning framework into a topology-modified generative adversarial network and mapping predefined business rules to the physical characteristics of fluids to construct a fluid dynamics model specifically includes: By combining topology-modification generative adversarial networks (GANs) with reinforcement learning frameworks, the model can continuously self-optimize during the learning process, and the quality of the generated results can be improved through the feedback mechanism of reinforcement learning. The generator in the topology-modification GAN is responsible for generating appropriate topologies based on the input spatial data, while the discriminator evaluates whether the generated topology conforms to preset rules and the characteristics of real data. With the reinforcement learning framework embedded, a reward mechanism guides the generator to optimize its output, and the generated topologies are strengthened and adjusted through continuous interaction. For example, by setting a reward function, the generator can automatically adjust parameters based on actual changes after each generation, gradually improving the rationality and accuracy of the generated topologies. Pre-set business rules (such as property rights regulations and building design standards related to real estate) are mapped to the physical characteristics of fluids, transforming the rules into physical quantities that can be understood by the computational model. For example, rules such as the functional use and area restrictions of buildings are transformed into parameters such as fluid velocity, pressure, or viscosity. These physical characteristics are reflected in the fluid dynamics model, which simulates the flow of fluid in space to represent the constraints and influences of business rules on spatial structures.

[0081] A pressure gradient field is generated using a fluid dynamics model, abrupt change points are identified using wavelet transform, and a side length threshold algorithm based on triangular mesh is used to obtain pressure abrupt change regions. Spatial heterogeneity features are extracted to generate a dynamic reward signal.

[0082] In this optional embodiment, a pressure gradient field is generated using a fluid dynamics model, abrupt change points are identified using wavelet transform, and pressure abrupt change regions are obtained using a triangular mesh-based side length threshold algorithm. Spatial heterogeneity features are extracted, and a dynamic reward signal is generated, including: The pressure field distribution is obtained by numerically simulating the target space using a fluid dynamics model, and the spatial gradient of the pressure field is calculated by the finite difference algorithm to generate the pressure gradient field. Wavelet transform was used to perform multi-scale analysis on the generated pressure gradient field, identify abrupt change points, and establish a cluster of abrupt change points. Construct a triangular network based on the established mutation point group, and draw a curve between the cumulative number of sides and the length of the triangle side. Calculate the slope of the initial line segment connecting the first and last points in the curve. The curves sorted by triangle side lengths are segmented using a differential algorithm. The slope of the line segment connecting the first and last points of each curve segment is compared with the slope of the initial line segment. A threshold is set based on the comparison results to filter out regions of sudden pressure changes. The pressure gradient field and pressure change region are analyzed to establish the physical relationship between fluid pressure and spatial topological changes, and spatial non-uniformity analysis is performed to extract spatial heterogeneity characteristics. The extracted spatial heterogeneity features are transformed into reward values ​​through nonlinear exponential mapping, and combined with random exploration incentive terms to generate dynamic reward signals.

[0083] It should be further explained that a curve is plotted with the cumulative number of sides as the x-axis and the triangle side length as the y-axis. The slope of the initial line segment reflects the spatial characteristics of the pressure change region. In addition, when the slope of a line segment is greater than the slope of the initial line segment, the side length value of the triangular mesh corresponding to the starting point of that segment is set as a threshold to filter out regions with pressure changes.

[0084] The generated dynamic reward signal is used as feedback input in the reinforcement learning framework to guide the generator to make topology corrections during adversarial training.

[0085] It should be added that when a large change in the pressure gradient field is detected in a certain area, the generator will correct the topology of that area based on the feedback of the reward signal to ensure that it conforms to the physical constraints of spatial topology changes. In this way, the generator can gradually optimize its generation strategy so that the final topology can meet the business rules while also achieving fluid dynamic rationality at the physical level.

[0086] The discriminator is used to evaluate the rationality of the topology correction results generated by the generator, and the evaluation results are fed back to the generator to optimize the topology correction parameters in order to generate a topology correction strategy.

[0087] It should be further explained that the discriminator evaluates the reasonableness of the topology correction results generated by the generator, and the evaluation results are fed back to the generator to optimize the topology correction parameters. Specifically, the topology correction strategy includes: A discriminator is used to evaluate the rationality of the topology correction results generated by the generator, determining whether the generated topology conforms to preset rules and physical constraints. For example, the discriminator can assess the rationality of the generated topology by analyzing factors such as the difference between the generated topology and the real data, the distribution of the pressure field, and the legality of the spatial structure. After completing the evaluation, the discriminator provides feedback based on the evaluation results. The feedback is usually a loss function value or reward value that measures the quality of the generated topology. This information is passed to the generator. After receiving the feedback, the generator further adjusts its topology correction parameters, such as the connection method between nodes, the length of edges, and the division of regions, to optimize the generated topology. For example, if the pressure gradient of the topology generated by the generator in some regions does not meet the physical requirements, the generator will adjust the topology of these regions based on the feedback to reduce pressure abrupt changes or unreasonable connections. Through continuous optimization and iteration, the generator will gradually learn the optimal topology correction strategy.

[0088] The graph structure update submodule is used to verify the topology correction strategy using incremental graph search technology, and combined with preset business weights and priorities, apply the verified topology correction strategy to the changed area to optimize the graph structure of the changed area.

[0089] The knowledge graph update module is used to update the real estate topology knowledge graph based on the optimized graph structure and the transaction characteristics of the graph database.

[0090] It should be further explained that, based on the optimized graph structure and the transaction characteristics of the graph database, updating the real estate topological knowledge graph specifically includes: The optimized graph structure reflects the improved results of the changed area after topological correction, including the reorganization and adjustment of nodes (such as land parcels and buildings) and edges (such as roads and connections). The graph database writes the optimized graph structure into the database through a transaction mechanism, while ensuring that each update can be traced and rolled back to ensure data integrity. Based on the optimized graph structure, the real estate topological knowledge graph is updated. The topological knowledge graph contains information such as relationships between real estate entities, spatial attributes, and functions, and is stored and managed through the graph database. When the topological structure changes, these changes need to be reflected in the knowledge graph in a timely manner, including updating the attributes of nodes (such as land use, building age, etc.) and updating the relationships between nodes (such as adjacency relationships, transportation connections, etc.).

[0091] The query results display unit 3 is used to construct a semantic parsing mechanism to convert the user's natural language query into the corresponding graph query language and display the query results using visualization tools.

[0092] It should be further explained that the semantic parsing mechanism, which transforms the user's natural language query into a corresponding graph query language and uses visualization tools to display the query results, specifically includes: A semantic parsing mechanism is constructed, using natural language processing (NLP) technology to transform user-input queries (such as "find commercial buildings located within a certain area") into machine-understandable graph query languages ​​(such as Cypher, SPARQL, etc.). Specifically, lexical and syntactic analysis is used to segment, tag, and analyze the syntactic structure of user input, identifying entities (such as "commercial buildings"), attributes (such as "area"), and relationships (such as "located in") within the query. Through semantic disambiguation and contextual understanding, the actual intent of the user's query is clarified, such as determining whether "area" refers to a geographical area or an administrative region. Based on the parsed semantic information, a corresponding graph query language is constructed for execution in the graph database. Using the generated graph query language, data in the graph database is queried to obtain node and edge information that meets the criteria. Visualization tools are then used to graphically display the obtained graph data. Nodes represent real estate entities or spatial objects, and edges represent relationships between entities (such as adjacency, inclusion, etc.). Through an interactive interface, users can click on nodes to view detailed information or dynamically adjust query conditions. This helps improve the query experience and data analysis capabilities of non-professional users, making data acquisition and decision support more efficient and accurate.

[0093] This invention employs "integrated image and text" technology to achieve image-text interaction between real estate spatial data and real estate registration information data. Through centralized data management, it establishes spatial databases and natural building entity models for land parcels, sub-districts, land parcels, and buildings, and constructs the correlation between land parcel, building, and housing data. Simultaneously, it builds an "integrated image and text" online query service system based on the Web to achieve "building search by image." By locating land parcels on the map, natural buildings can be found, and the floor, unit, and unit number of the building can be displayed intuitively in the form of a building list. This achieves integrated query and information display of land, building, and housing information, providing the public with powerful visual query functions, allowing the public to more conveniently and intuitively understand the detailed information of land parcels and houses.

[0094] Data synchronization unit 4 is used to acquire real estate business operations and convert them into events in a predefined format. Event transmission is performed through a distributed message queue, and real-time processing is carried out in combination with streaming topology technology to ensure the consistency between the real estate knowledge graph and the graphical data.

[0095] It should be further explained that the process of acquiring real estate business operations and converting them into events in a predefined format, transmitting these events via a distributed message queue, and processing them in real time using streaming topology technology, specifically includes: By capturing real estate business operations, such as transactions, registrations, and changes, these are transformed into events in a predefined format. Each event contains key information about the business operation (such as transaction amount, change details, and related land parcels) and is standardized according to a specific format (such as JSON or XML). Distributed message queues (such as Kafka and RabbitMQ) are used to transmit these events, ensuring efficient and reliable data transfer between multiple systems or services. Combined with streaming topology technology, the transmitted events are processed in real time. The event stream in the message queue is analyzed and processed, and each event is treated as a data stream for real-time computation, updating the system status promptly. Through streaming processing, the system can immediately respond to changes in real estate operations, such as updating node attributes or relationships in the real estate knowledge graph based on new transaction information, or dynamically adjusting graphical data based on registration changes. This ensures that data is not lost or erroneous during changes, providing decision-makers with accurate and real-time data information to support precise spatial analysis and business decisions.

[0096] In a specific embodiment, such as Figure 3 As shown, this invention adopts a service-oriented architecture as the component model, encapsulating different functional units of the application as services, and realizing the interaction between services through clearly defined interfaces and contracts. These interfaces are defined in a neutral and standard manner, independent of the software and hardware platforms, operating systems and programming languages ​​on which the services depend, thereby ensuring that services in various heterogeneous systems can communicate and integrate in a unified and universal way. This architecture supports integrated text and image query and visualization of spatial and attribute data such as land parcels, natural buildings, and houses, improving the interoperability and data sharing capabilities of the system.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time query system for real estate registration information based on graphic data, characterized in that, The system includes: The knowledge graph generation unit is used to process the established multimodal dataset, generate joint vectors, encode and store the joint vectors in a preset graph database, and simultaneously build a dual dynamic indexing mechanism to generate a real estate knowledge graph. The knowledge graph update unit is used to capture change events in the multimodal dataset in real time. By locating the changed area and its neighborhood, it identifies the graph structure differences in the changed area and generates a topology correction strategy. It then updates the real estate topology knowledge graph by combining the transaction characteristics of the graph database. The query results display unit is used to build a semantic parsing mechanism to convert users' natural language queries into corresponding graph query languages ​​and display the query results using visualization tools. The data synchronization unit is used to acquire real estate business operations and convert them into events in a predefined format. Event transmission is performed through a distributed message queue, and real-time processing is combined with streaming topology technology to ensure the consistency between the real estate knowledge graph and the graphical data.

2. The real-time query system for real estate registration information based on graphic data according to claim 1, characterized in that, The query results display unit also includes a rendering module; used to obtain the reference coordinates of the current rendering interface and extract the bounding box of the rectangle under the current rendering interface based on the R* tree space index. In response to the selected attribute information, and based on the bounding box of the current rendering interface, update the rendering mapping information of each node within it. In response to the user's movement of the rendering interface, a graphics data matrix is ​​used as the underlying logic for switching rendering, and the rendering mapping information of the nodes is superimposed on the graphics data matrix; where the graphics data matrix is ​​the rendering interface before the selected attribute information.

3. The real-time query system for real estate registration information based on graphic data according to claim 2, characterized in that, Updating the rendering mapping information of each node specifically includes: the rendering mapping information is based on the HSV model and a hue reference vector is pre-generated; the first attribute information under the current rendering interface and the corresponding saturation vector are obtained; in response to the selection of the second attribute information, the corresponding saturation difference is found in the hue reference matrix and superimposed on the saturation vector to obtain the target vector; the minimum distinguishing value, the target vector, and the target vector are substituted into the dynamic programming algorithm to obtain the saturation vector corresponding to the second attribute information.

4. The real-time query system for real estate registration information based on graphic data according to claim 1, characterized in that, The knowledge graph generation unit includes: The dataset creation module is used to collect graphic data and related attribute information of real estate from multi-source heterogeneous platforms, and combine spatiotemporal benchmark dynamic alignment algorithm and blockchain-based spatiotemporal anchoring mechanism to create a multimodal dataset. The joint vector generation module is used to construct a dual-channel adversarial generative network that includes geometric and semantic channels, and to perform cross-modal adversarial training on the established multimodal dataset to generate joint vectors. The storage and construction module is used to encode joint vectors into composite semantic codes using a spatiotemporal-aware encoder, and write the semantic codes into a preset graph database, while simultaneously constructing a dual dynamic indexing mechanism of R* tree spatial index and semantic inverted index. The knowledge graph building module is used to analyze the semantic similarity and topological relationships between entity nodes in real time based on the completed graph database and the constructed dual dynamic indexing mechanism, and to build a real estate knowledge graph.

5. A real-time query system for real estate registration information based on graphic data according to claim 4, characterized in that, The storage and construction module includes: The composite semantic coding submodule is used to reconstruct the graph attention mechanism using a pre-trained spatiotemporal awareness encoder, design a multi-layer attention network, generate a fused feature vector, and map the fused feature vector to a composite semantic code through a multi-layer perceptron. The storage submodule is used to select the native stream processing pipeline of the graph database and write the generated composite semantic code and the corresponding multimodal dataset as batch transactions into the preset graph database for storage. The index construction submodule is used to construct an R* tree spatial index based on the spatial grid features in the composite semantic encoding, and to parse the entity type code and relation topology code of the composite semantic encoding to construct a semantic inverted index. The indexing mechanism submodule is used to build a dual dynamic indexing mechanism by sharing storage pointers of composite semantic encoding, so as to perform real-time association and collaborative access between the R* tree space index and the semantic inverted index.

6. A real-time query system for real estate registration information based on graphic data according to claim 5, characterized in that, The composite semantic encoding submodule includes: By combining a pre-trained spatiotemporal-aware encoder with a spatial relation distillation mechanism, spatial topological prior vectors are generated, and the graph attention mechanism is reconstructed by combining preset rule constraints. Based on the reconstructed graph attention mechanism, a multi-layer attention network is designed, which includes a spatial grid layer, a temporal slice layer, an entity type layer, and a relational topology layer. The joint vector is input into a multi-layer attention network. By calculating the attention weight of each layer, the corresponding feature vector is output. The feature vectors are then weighted and fused together with the weight matrix of each layer to generate a fused feature vector. By using a multilayer perceptron to perform nonlinear mapping on the fused feature vectors, a composite semantic code expressing spatial location, temporal context, entity category, and topological semantics is generated.

7. A real-time query system for real estate registration information based on graphic data according to claim 6, characterized in that, The process of generating spatial topology prior vectors by combining a pre-trained spatiotemporal-aware encoder with a spatial relation distillation mechanism, and reconstructing the graph attention mechanism by combining preset rule constraints, includes: extracting topological features at different spatial scales using a hierarchical graph convolutional network based on the pre-trained spatiotemporal-aware encoder, and distilling and abstracting the topological features to generate spatial topology prior vectors; transforming preset rule constraints into rule damage functions, and constructing a rule constraint engine containing spatial topological information and rule semantic relationships based on the generated spatial topology prior vectors; and reconstructing the graph attention mechanism based on the graph attention mechanism in the spatiotemporal-aware encoder using the topology prior terms and rule bias terms derived from the rule constraint engine.

8. A real-time query system for real estate registration information based on graphic data according to claim 1, characterized in that, The knowledge graph update unit includes: The change event capture module is used to build an event listening pipeline using message flow technology to capture change events of graphical data in multimodal datasets in real time. The change area positioning module is used to locate the change area and its surrounding range using multi-level spatial indexing technology based on the coordinate information of the change event. The graph structure optimization module is used to analyze and identify graph structure differences in changed areas, formulate topology correction strategies using topology correction technology, and optimize the graph structure of changed areas by combining preset business weights and priorities. The knowledge graph update module is used to update the real estate topology knowledge graph based on the optimized graph structure and the transaction characteristics of the graph database.

9. A real-time query system for real estate registration information based on graphic data according to claim 8, characterized in that, The graph structure optimization module includes: The target node pair filtering submodule is used to construct a local subgraph based on the located change area and neighborhood range, and to filter out target node pairs associated with the change area through spatial indexing technology. The graph structure difference identification submodule is used to perform topological relationship calculations on the selected target node pairs using incremental graph algorithms to identify graph structure differences in changed areas. The topology correction policy submodule is used to construct a topology correction generative adversarial network, which combines a reinforcement learning framework with stress field analysis techniques to generate dynamic reward signals and generate topology correction policies based on the dynamic reward signals. The graph structure update submodule is used to verify the topology correction strategy using incremental graph search technology, and combined with preset business weights and priorities, apply the verified topology correction strategy to the changed area to optimize the graph structure of the changed area.

10. A real-time query system for real estate registration information based on graphic data according to claim 9, characterized in that, The topology correction strategy submodule includes: constructing a topology correction generative adversarial network containing a generator and a discriminator based on the identified graph structure differences; mapping preset business rules to the physical characteristics of fluids using a reinforcement learning framework, generating dynamic reward signals through pressure field analysis technology to guide the generator to perform topology correction; using the discriminator to evaluate the rationality of the topology correction results generated by the generator, and feeding the evaluation results back to the generator to optimize the topology correction parameters to generate a topology correction strategy.

11. A real-time query system for real estate registration information based on graphic data according to claim 9, characterized in that, The method of combining a reinforcement learning framework to map preset business rules to the physical characteristics of fluids and generating dynamic reward signals through pressure field analysis to guide the generator in topology correction includes: embedding a reinforcement learning framework into a topology correction generative adversarial network, mapping preset business rules to the physical characteristics of fluids, and constructing a fluid dynamics model; using the fluid dynamics model to generate a pressure gradient field, using wavelet transform to identify abrupt change points, and using a triangulation-based side length threshold algorithm to obtain pressure abrupt change regions, extracting spatial heterogeneity features, and generating a dynamic reward signal; and using the generated dynamic reward signal as feedback input in the reinforcement learning framework to guide the generator in topology correction during adversarial training.

12. A real-time query system for real estate registration information based on graphic data according to claim 10, characterized in that, The process of generating a pressure gradient field using a fluid dynamics model, identifying abrupt change points using wavelet transform, and obtaining pressure abrupt change regions using a triangular mesh-based side length thresholding algorithm to extract spatial heterogeneity features and generate a dynamic reward signal includes: numerically simulating the target space using a fluid dynamics model to obtain the pressure field distribution, calculating the spatial gradient of the pressure field using a finite difference algorithm to generate a pressure gradient field; performing multi-scale analysis on the generated pressure gradient field using wavelet transform to identify abrupt change points and establish a cluster of abrupt change points; constructing a triangular mesh based on the established cluster of abrupt change points, and plotting the relationship between the cumulative number of sides and the triangular side lengths. The curve is plotted, and the initial slope of the line segment connecting the first and last points is calculated. A differential algorithm is used to segment the curve based on the triangular side lengths. The slope of the line segment connecting the first and last points of each segment is compared with the initial slope. A threshold is set based on the comparison results to filter out pressure abrupt change regions. The pressure gradient field and pressure abrupt change regions are analyzed to establish the physical relationship between fluid pressure and spatial topological changes. Spatial non-uniformity analysis is performed to extract spatial heterogeneity features. The extracted spatial heterogeneity features are converted into reward values ​​through nonlinear exponential mapping and combined with random exploration incentive terms to generate a dynamic reward signal.

Images